QFM115: Engineering Leadership Reading List - May 2026

The case here is that software has always been chaotic and barely understood, whatever the clean-profession image suggests -- and that the AI panic, real as it is, distracts from the older truth that nobody in tech quite knows what they are doing. It runs on one sustained metaphor: commanding a ship with broken instruments, missing manuals and a confused crew, where projects move by improvisation rather than any competent plan. Set against that, the fear of AI taking the jobs starts to look almost quaint.

Models are genuinely good at programming now, but what you get out depends almost entirely on what you bring: a strong developer like Matt Perry compounds existing domain knowledge into big gains, while a beginner stalls past the first prototype because the model has no holistic sense of architecture. The point isn't that AI replaces developers -- it is an instrument, and the same instrument sounds very different in a master's hands and an amateur's.

AI earns its keep more by raising code quality than by raising output speed. Run several models -- Claude, Codex, Cursor -- over a pull request, rank what they flag by severity, fix the critical and high issues and bin the false positives, and you surface pre-existing bugs and weak spots even when none of it makes you faster. It turns the model from a 'slop cannon' that emits barely-passable code into a quality tool that catches the subtle stuff.